Anthropic's '2028' essay on US-China AI competition fractures the industry — Nvidia, HuggingFace, and OpenAI each offer incompatible counter-doctrines.
Anthropic's '2028' essay frames the AI finish line as recursive self-improvement, splitting the industry on compute restriction vs open-source export strategy.
DeepSeek v4 Flash Thinking beats Gemini 3.1 Flash Lite on a scientific reasoning benchmark in all three rounds, including self-verification stability.
DeepSeek raising $3–4B at $50B valuation via China's national AI fund shifts the open-weights frontier to state-backed capital.
subQ claims 12M-token context at 52× FlashAttention speed, but benchmarks test only the 1M preview model, with figures differing between video and website.
After Claude rate limits, a researcher ran 10M+ tokens on DeepSeek V4 and was shocked by the cost gap. swyx: 'Efficiency is back on the menu again boys.'
DeepSeek's Visual Primitives paper uses coordinate tokens in chain-of-thought to achieve ~10× KV-cache compression vs. Sonnet 4.6 and Gemini 3 Flash on vision tasks.
NIST CAISI evaluation: DeepSeek V4 Pro is China's most capable AI model but lags leading US models by roughly eight months, per official US government benchmarking.
NIST CAISI confirms DeepSeek V4 Pro as top Chinese model, ~8 months behind US leaders — but MIT-licensed, 1M context, and 50–100× cheaper than closed rivals.
DeepSeek V4 is out: 1.6T open-source parameters, 1M token context, 3.7× fewer FLOPs than V3.2, scoring 120/120 on Putnam 2025.
Matthew Berman argues DeepSeek confirms US open-source AI has no business model without government subsidy; proposes compute quotas and sovereign procurement as policy fixes.
DeepSeek v4 reignites debate on US open-source AI: Berman argues the business model is broken, leaving Nvidia as the only credible US champion.
Developers running DeepSeek V4 Flash with 2-bit selective GGUF via llama.cpp describe it as 'the first time I feel I have a frontier model running on my computer' — a milestone for local AI.
Intel releases W4A16 INT4 quantizations of DeepSeek-V4-Pro and Flash via AutoRound — no MXFP4 hardware required, expanding which hardware can self-host DeepSeek V4 at near-full quality.
DS2API exposes DeepSeek Web as OpenAI/Claude/Gemini wire-compatible APIs — reverse-engineered, disclaimer-heavy, but a clear signal of demand for free API access to frontier-adjacent models.
DeepSeek-V4's MIT-licensed 1M-context MoE and Kimi-K2.6's multimodal orchestration create the first complete open-weights agentic deployment stack.
DeepSeek V4 drops two open-weight models with 1M-context by default, CSA+HCA hybrid attention, and V4-Pro priced at roughly 1/7 Opus 4.7's output cost.
DeepSeek V4's 10× KV-cache compression restructures AI cost economics globally, exposing a structural threat to US lab pricing and strategic positioning.
US director Michael Kratzios confirms foreign distillation attack campaigns on US AI labs; Anthropic data shows DeepSeek had far fewer exchanges than suspected.
DeepSeek V4-Pro open-sourced with 1.6T params, 1M context window, and 10x KV cache reduction vs V3.2 — #1 HuggingFace trending in 43 minutes.
Huawei confirms its Ascend 950-based supernode will fully support DeepSeek V4, completing China's domestic frontier AI hardware-software stack.
DeepSeek V4-Pro launches with 1.6T parameters, 1M context, and 10× KV cache reduction over V3.2 — multiplying inference concurrency roughly 10× on the same hardware.
Curated AI insights — sent when there's something worth your inbox.